Results

Participant Characteristics

We enrolled 671 patients from 3 ICUs and excluded 174 patients. The final cohort consisted of 497 patients: 137 (27.6%) received no aerosol inhalation, 262 (52.7%) received glucocorticoid aerosol, and 98 (19.7%) received aerosol inhalation without glucocorticoid (Figure 1). We isolated A. baumannii from 159 (32.0%) patients. The median patient age was 60.1 (IQR 49–73) years, and 67.8% were male. The median length of ICU stay was 15 (IQR 7–23) days. Besides A. baumannii, the 3 other bacteria commonly isolated were Klebsiella pneumoniae (n = 38, 7.6%), P. aeruginosa (n = 27, 5.4%), and E. coli (n = 15, 3.0%) (Appendix Table 1). Most (22.5%) study patients received glucocorticoid therapy for ARDS and for asthma or COPD (21.3%) (Appendix Table 2).

Enlarge

Figure 1. Flowchart for enrolling participants in a study of Acinetobacter baumannii among patients receiving glucocorticoid aerosol therapy during invasive mechanical ventilation, China. ICU, intensive care units.

Risk Factors for A. baumannii Isolation

We performed univariate Cox regression analysis of risk factors for A. baumannii isolation (Table 1). Compared with no aerosol inhalation, glucocorticoid aerosol therapy had a statistically significant effect on A. baumannii isolation (HR 1.860, 95% CI 1.264–2.738; p = 0.002). Aerosol inhalation without glucocorticoid was not a risk factor for A. baumannii (p>0.05). Other candidate risk factors were cardiovascular diseases, chronic renal insufficiency, COPD or asthma, current or former smoking history, use of broad-spectrum antimicrobial drugs for ≥7 days, invasive mechanical ventilation for ≥5 days, vasopressor treatment, renal dialysis for ≥3 days, and APACHE II score.

To assess whether glucocorticoid aerosol therapy was an independent risk factor for A. baumannii isolation, we established 2 models using multivariate Cox regression analysis in complete cases. Model 1 included all variables, and model 2 only included variables with p<0.1 in the univariate analysis. Glucocorticoid aerosol was an independent risk factor for A. baumannii isolation in both model 1 (HR 1.499, 95% CI 1.001–2.246; p = 0.049) (Appendix Table 3) and model 2 (HR 1.528, 95% CI 1.024–2.278; p = 0.038) (Table 2). Cardiovascular diseases, prolonged use of broad-spectrum antimicrobial drugs, invasive mechanical ventilation, and vasopressor treatment were other independent risk factors for A. baumannii (Table 2; Appendix Table 3). As a whole variable, aerosol inhalation had no effect on A. baumannii isolation (Appendix Table 4).

In the propensity-matched cohort, the possible glucocorticoid-related covariables were balanced in both groups (Appendix Table 5). Univariate and multivariate Cox regression analyses also indicated that glucocorticoid aerosol therapy was an independent risk factor for A. baumannii isolation (Table 3; Appendix Table 6). In an independent model that included indications for glucocorticoid aerosol therapy, risk factors for A. baumannii isolation were glucocorticoid aerosol treatments for COPD or asthma and for ARDS (Appendix Table 7).

Log-rank analysis showed that the difference among the groups was statistically significant (p<0.001). The cumulative hazard for A. baumannii isolation was significantly higher in the glucocorticoid aerosol group compared with the no aerosol inhalation (HR 1.871; 95% CI 1.206–2.772; p<0.001) and aerosol inhalation without glucocorticoid (HR 2.316; 95% CI 1.482–3.620; p = 0.002) groups (Figure 2).

Enlarge

Figure 2. Kaplan-Meier curves of cumulative hazards of different aerosol inhalation treatments on Acinetobacter baumannii isolation among patients during invasive mechanical ventilation, China. We used log-rank analysis to compare hazard ratios over time among all groups (p<0.001), glucocorticoid aerosol therapy group with the no aerosol inhalation group (p<0.001), and the glucocorticoid aerosol therapy group with the aerosol inhalation without glucocorticoid group (p = 0.002).

Effect of Glucocorticoid Aerosol Therapy on A. baumannii Isolation among Subgroups

We divided patients into subgroups to evaluate the contribution of glucocorticoid aerosol to A. baumannii isolation from different subpopulations. We found glucocorticoid aerosol was a promoting factor for A. baumannii isolation from most subpopulations, except patients with type 2 diabetes mellitus, hematologic malignancy, antimicrobial drug use for A. baumannii, and short ICU stays (p>0.05) (Figure 3). We noted no statistically significant interactions between most prespecified subgroups defined by demographics, medical history, underlying conditions, APACHE II score, treatment measures, and length of ICU stay (interaction p>0.05). The favorable effect of glucocorticoid aerosol on A. baumannii isolation was relatively greater in the subgroup of patients with longer vasopressor treatment (interaction p = 0.006) (Figure 3).

Enlarge

Figure 3. Effects of glucocorticoid aerosol therapy on Acinetobacter baumannii isolation during invasive mechanical ventilation among prespecified patient subgroups, China. We applied Cox proportional hazards model with Efron’s method for handling ties between groups to assess favorable effects of glucocorticoid aerosol on A. baumannii isolation. Black horizontal marks indicate hazard ratios, and error bars indicate 95% CIs. APACHE II, Acute Physiology and Chronic Health Evaluation II; COPD, chronic obstructive pulmonary disease; ICU, intensive care unit. *Limited numbers for analysis. †For interaction, p = 0.006.

Association between Glucocorticoid Aerosol Therapy and Clinical Prognosis

We performed univariate and multivariate Cox regression analyses to evaluate the possible risk factors for 30-day mortality in critically ill patients on ventilators. We found glucocorticoid aerosol was not a risk factor for 30-day mortality in those patients, but A. baumannii isolation was independently associated with 30-day mortality (HR 1.824, 95% CI 1.317–2.104; p = 0.045) (Appendix Table 8,9). A further separate analysis of A. baumannii isolation status showed that A. baumannii infection was independently associated with 30-day mortality (HR 2.759, 95% CI 1.575–4.833; p = 0.012) (Appendix Table 8, 10).